Aquaculture Fish Disease Detection Using a Hybrid Dual-Branch Convolutional Neural Network

Aquaculture has been widely used in the world in terms of its contribution to the global food security but fish diseases have been a major challenge to sustainable production. These diseases need to be detected early and accurately to reduce the massive death rates and financial losses. Deep learning-based methods of image classification in fish disease detection are investigated in this paper. The dataset employed to test four models, namely ResNet50, EfficientNetB0, MobileNetV2, and the proposed Hybrid Dual-Branch CNN, combining ResNet50 with MobileNetV2, consisted of 697 images of seven fish diseases. According to the experimental outcomes, ResNet50 and EfficientNetB0 demonstrated the highest accuracy of 95.55% and 94.26, respectively, and MobileNetV2 had the least accuracy of 66.00. The proposed Hybrid Dual-Branch CNN performed better than all the other models in the two-class classification scenario with an accuracy of 96.84 and a macro-F1 score of 0.97. This enhanced performance is due to the complementary nature of feature extraction in the two networks where ResNet50 is more efficient in deep structural and textural features and MobileNetV2 is specialized in lightweight shape and colour patterns. These results highlight the promising nature of hybrid architectures in strong and scalable fish disease diagnostics in aquaculture.